wigest: a ubiquitous wifi-based gesture recognition system · wigest: a ubiquitous wifi-based...

42
WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser†, Moustafa Youssef ‡, and Khaled A. Harras* Alexandria Univ.†, Egypt-Japan Univ. of Sc. and Tech.‡, Carnegie Mellon Univ.*

Upload: others

Post on 08-Sep-2020

6 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

WiGest: A Ubiquitous WiFi-based Gesture Recognition System

Heba Abdelnasser†, Moustafa Youssef‡, and Khaled A. Harras*Alexandria Univ.†, Egypt-Japan Univ. of Sc. and Tech.‡, Carnegie Mellon Univ.*

Page 2: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Motivation

● Rendering touch input – user is wearing gloves.

● Wet/dirty hands.

Page 3: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Motivation

● Rendering touch input – user is wearing gloves.

● Wet/dirty hands.

● Special sensors● E.g. Samsung Galaxy S5

Page 4: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

WiGest Approach

● Free the user from specialized sensors● Using the ubiquitous WiFi● Works with any phone

● Leveraging natural human movements to control devices

Page 5: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Basic Idea

• Leveraging the impact of hand motion on the received WiFion the phone to control WiFi-enabled devices.

Page 6: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

WiGest Example

Page 7: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Challenges

• WiFi is a noisy signal• Avoiding false positives – human

interference• Handling the variations of

gestures and their attributes

• Energy consumption• No training

– Handle different humans– Same human a different times

-36

-33

-30

-27

-24

1 2 3 4 5 6 7 8

RS

SI

dB

m

Time in Seconds

-41

-40

-39

-38

-37

-36

-35

-34

0 1 2 3 4 5 6 7 8 9 10

RS

SI

dB

m

Time in Seconds

Far

Near

Page 8: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Agenda

Introduction• WiGest System• Results• Conlusions

Page 9: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Agenda

Introduction• WiGest System• Results• Conlusions

Page 10: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

WiGest System

Page 11: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

WiGest System

Page 12: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Primitive Extraction

Page 13: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Primitive Extraction

1. Denoising– Using Wavelet denoising

2. Edges detection,– Extract different primitives.

3. Parameters extraction,– Magnitude: near/far.– Speed: fast/slow.

-36

-33

-30

-27

-24

1 2 3 4 5 6 7 8

RS

SI

dB

m

Time in Seconds

Input Output

Page 14: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Noise Reduction

Time in sec

Scale

2.5 3 3.5 4 4.5 5 5.5 6 6.5 7

5

10

15

20

25

30

35

-36

-33

-30

-27

-24

1 2 3 4 5 6 7 8

RS

SI

dB

m

Time in Seconds

Page 15: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Noise Reduction

• Removing noise using wavelet denoising– It can preserve signal details while filtering out the noise

and variations in the signal– Consists of three stages:

• Decomposition the signal to approximation and detail coefficients.

• Thresholding detail coefficients• Reconstruct the denoised signal by adding approximation and

detail after thresholding

Page 16: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Noise Reduction

Raw signal After denoising

-33

-30

-27

-24

1 2 3 4 5 6 7 8

RS

SI

dB

m

Time in Seconds

-36

-33

-30

-27

-24

1 2 3 4 5 6 7 8

RS

SI

dB

m

Time in Seconds

Page 17: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Edges Extraction

• Detecting gesture edges using wavelet analysis– In detail (high-pass filter) coefficients

• Falling edge causes a local maxima• Rising edge causes a local minima

Page 18: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Edges Extraction

Denoised signal

DWT detail coefficient

-33

-30

-27

-24

1 2 3 4 5 6 7 8

RS

SI

dB

m

Time in Seconds

-6

-4

-2

0

2

4

6

1 2 3 4 5 6 7 8

Time in Seconds

Page 19: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Edges Extraction

Denoised signal

DWT detail coefficient

-33

-30

-27

-24

1 2 3 4 5 6 7 8

RS

SI

dB

m

Time in Seconds

Page 20: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Edges Extraction

Denoised signal

DWT detail coefficient

Page 21: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Edges Extraction

Denoised signal

DWT detail coefficientExtracted primitives

Page 22: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Parameters Extraction

Magnitude

-41

-40

-39

-38

-37

-36

-35

-34

0 1 2 3 4 5 6 7 8 9 10

RS

SI

dB

m

Time in Seconds

Far

Near

-39

-38

-37

-36

-35

0 2 4 6 8 10 12 14 16 18

RS

SI

dB

m

Time in Seconds

Slow Fast

Speed

Page 23: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Parameters Extraction

Magnitude

-41

-40

-39

-38

-37

-36

-35

-34

0 1 2 3 4 5 6 7 8 9 10

RS

SI

dB

m

Time in Seconds

Far

Near

-39

-38

-37

-36

-35

0 2 4 6 8 10 12 14 16 18

RS

SI

dB

m

Time in Seconds

Slow Fast

Speed

Page 24: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

System Logic Flow

Page 25: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Segmentation

• Silence period determines the start and end of a gesture

• Preamble (unique signature), to start the communication channel

Frequency =

3 /2 Hz,

Count = 2

Frequency =

5 /2 Hz,

Count = 3

-39

-36

-33

-30

0 1 2 3 4 5 6 7 8

RSSI

dBm

Time in SecondsSilence period Gesture delimiter

Page 26: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Pattern Encoding

• The extracted gesture primitives are converted to string sequence:

– Rising edge : +– Falling edge : -– Pause : 0

• Gesture matching– String matching to identify the gesture

Extracted gesture pattern : - + - +Gesture : Infinity

Page 27: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Gesture Identification

Extracted gesture pattern : - + - +Gesture : Infinity

Preamble • Energy efficiency• Human interference

and noise

Pattern encoding• Efficient matching• Error tolerance

Page 28: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Gestures ParametersFrequency =

3 /2 Hz,

Count = 2

Frequency =

5 /2 Hz,

Count = 3

-39

-36

-33

-30

0 1 2 3 4 5 6 7 8

RSSI

dBm

Time in Seconds

Count• Number of gesture

repetitions

• E.g. double click and single click

Frequency• Number of gesture

repetitions per unit time

• E.g. speed of basketball

dribble

Page 29: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

System Logic Flow

Page 30: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Action Mapping

• At the end, the extracted gesture is mapped to an

application action .

Gesture Action

Application

Page 31: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Agenda

IntroductionWiGest System• Results• Conlusions

Page 32: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Implementation and Evaluation

• Test environments• Typical apartment• Engineering building at our campus

• More than 1000 experiment• Off-the-shelf WiFi-equipped devices• Hardware:

• Cisco Linksys

• Android-based cell phone.

• HP laptop

39 ft

34 ft

115 ft

49 ft

Page 33: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Experiments Performed

• Distance effect for different scenarios• Orientation• Number of APs• Human interference

• Whole home case study

• More results in the paper

Page 34: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Distance Impact

0.5

0.6

0.7

0.8

0.9

1

3 7 10 13 16 20 23 26 39 52 62-70

-60

-50

-40

-30

Acc

ura

cy

Av

erag

e R

SS

I

Distance in feet

Accuracy Average RSSI

0.5

0.6

0.7

0.8

0.9

1

3 7 10 15 20 26 30 39 56-70

-60

-50

-40

-30

Acc

ura

cy

Av

erag

e R

SS

I

Distance in feet

Accuracy Average RSSI

0.5

0.6

0.7

0.8

0.9

1

7 10 13 16 20 30 36-70

-60

-50

-40

-30

Acc

ura

cy

Av

erag

e R

SS

I

Distance in feet

Accuracy Average RSSI

• 87% accuracy for distances up to 26ft

• Using a single AP

Page 35: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Orientation Impact

• Overall accuracy of 90.5%• The highest accuracy is in West – body not blocking the

signal

0

0.2

0.4

0.6

0.8

1

North South East West Overall

Acc

ura

cy

Page 36: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Number of APs Impact

0

0.2

0.4

0.6

0.8

1

1 APs 3 APs 5 APs 7 APs

Acc

ura

cy

Number of APs used in voting

• Accuracy 96% using 3 APs

• Reaches 100% with seven APs

Page 37: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Human interference

• Accuracy 89% in the presence of four

interfering humans

0.5

0.6

0.7

0.8

0.9

1

1 2 3 4

Det

ecti

on

Acc

ura

cy

Number of interfering humans

Page 38: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Whole Home Media-player Case Study

• Accuracy of 96% using 2 APs

0 .9

0 .96

0 .93

0 .97

0 .99

1

0 .98 0.02

0.01

0.025

0.03

0.04

0.090.005 0.005

0.005

0.005

0.03 0.005

0

0 0 00

0 0000

0 0000

00 0000

0000

000

00

0

0

0

00

00

nu

ll

Ac

tu

al

Ac

tio

n P

erfo

rm

ed

Act ion Classif ied

39 ft

34 ft

Page 39: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Agenda

IntroductionWiGest SystemResults• Conlusions

Page 40: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Conclusion

• WiGest is a gesture recognition system• Robust to noise in the environment and interfering humans• Does not require any training• Energy-efficient• Works with any WiFi-enabled device

• Primitives detection accuracy is 87.5%– Using a single AP

• Accuracy increases to 96% using three overheard APs.

Page 41: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Future Work

• Leveraging CSI for finer grained gestures• Other wireless technologies, e.g. cellular and

Bluetooth• Other applications

Page 42: WiGest: A Ubiquitous WiFi-based Gesture Recognition System · WiGest: A Ubiquitous WiFi-based Gesture Recognition System Heba Abdelnasser, Moustafa Youssef, and Khaled A. Harras*

Thank You

Come and see our demo tomorrow10am to 1pm

wrc-ejust.org wrc_ejust wrc.ejust